Skip to main content

​
Overview

The prompt script is the set of instructions that tells your AI agent how to behave. It controls what the agent says, what questions it asks, and when it should book an appointment. Think of it as a sales script for your AI.

​
The prompt editor

Navigate to your agent’s Settings → Prompt to access the editor. The editor supports:
  • Free-form text editing (up to 50,000 characters)
  • Version management with a dropdown selector
  • Diff indicator showing changes vs. the active version (+N / -M vs. active vK)
  • Missing integration warnings when your prompt references a calendar that isn’t connected

​
Version management

SetterMaxer uses a draft/publish model for prompt scripts:
1

Edit a draft

All changes are saved as drafts. Drafts are editable and deletable. You can have multiple drafts per agent.
2

Publish

When you’re ready, publish a draft to make it the active version. Published versions are immutable — they can’t be edited or deleted.
3

Restore

You can restore any previously published version to make it active again without creating a new version.

​
Save actions

​
AI prompt generation

You can regenerate a prompt at any time:
  1. Click Create new prompt in the prompt settings
  2. Enter your company name and product description
  3. Optionally click Prefill from website to scrape details from a URL
  4. Choose your integration type
  5. The AI generates a new draft based on your input

​
The playground

The Playground page provides a split-pane layout — prompt editor on the left, chat on the right — so you can iterate on your prompt and immediately test it.
  • Edit the prompt and save, then the chat resets for a fresh test
  • The test conversation uses the version you’re currently editing
  • A “Re-run last message” button lets you quickly retry after saving changes
Use the playground to fine-tune your agent’s responses before publishing. Changes in the playground are saved as drafts — they don’t affect live conversations until you publish.

​
Unsaved changes guard

If you have unsaved edits, SetterMaxer blocks navigation and version switching with a confirmation modal. This prevents accidentally losing work when switching between versions or leaving the page.

​
Information to capture

Below the prompt editor you can name information you want your agent to pick up from its conversations — a fitness goal, a budget, a preferred location. Each one needs two things:
  • Name — what you call it, e.g. Fitness goal. This is the column heading in your contact list.
  • What should the agent capture? — a plain sentence, e.g. “the lead’s main fitness goal”.
Each detail becomes a column in your contact list, so you can filter and sort by it, and it is sent to your webhooks and Zaps. Your agent records these as they come up naturally, and only stores what the lead actually said. It never asks a question purely to fill one in — so if you want something filled in reliably, your prompt script has to ask for it. Anything nobody mentions simply stays empty.
Each detail also gets a field name (e.g. fitness_goal), filled in for you from the name you chose. That is what webhooks and Zapier use, so it is fixed once the detail is created — renaming the detail later will not break your integrations. Use Change while creating it if you want a different one.
If you edit a captured value yourself — in the contact list, by import, or through Zapier — your agent stops changing it, so your correction stands. Use Let the agent fill this again on the contact to hand it back. Removing a detail asks what you mean: Stop capturing it leaves the column and its values in your contact list, while Delete the column removes it from there too.

​
Agent settings on the prompt page

The prompt page also includes: Automatic follow-ups have their own page: see Follow-up messages.

​
Split long replies into multiple messages

With this on, the agent may answer in two or three short messages instead of one long block — the way a person texts. The agent decides where to break, so short answers, greetings and single questions still arrive as one message. The parts appear one after another on the website widget and in the playground, and are sent as separate WhatsApp messages a moment apart. Turn it off to always send one message per reply.